9.4. Limitations and conclusion
All data in this study were self-reported, increasing the potential for common-method bias (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). However, we expect that common method variance did not significantly influence our findings for several reasons. The independent and dependent variables were gathered at two different points in time, more than a decade apart. In addition, many of the study's constructs were demographic (e.g., gender) or biographical (e.g., existence of a mentoring relationship, organizational position), as opposed to perceptual in nature. Moreover, the MSCEIT is an ability-based test with correct answers, making it difficult to provide false or socially desirable answers.
While our response rate aligned with norms in social science research, our sample size was relatively small and homogeneous. We had adequate numbers to test our hypotheses, but our analyses were limited in the number and types of controls we could include and the types of analyses we could perform. Our respondents were primarily Caucasian, which precluded analyzing the effects of race and ethnicity, and almost all were undergraduate business majors, which narrowed the range of potential career choices. Moreover, our sample contained a higher percentage of males (66.7% versus 53.9%) compared to the gender composition of the Time 1 sample and tended to have slightly higher undergraduate GPAs (higher means and reduced variance). These attrition effects were relatively small in magnitude and consistent with both: 1) prior research showing that subjects higher in mental abilities are more likely to remain in longitudinal studies (Baltes et al., 1971; Goodman & Blum, 1996) and 2) demographic trends regarding women's participation in the workforce (Lu et al., 2017). Nevertheless, additional and larger longitudinal data sets that track the evolution of career paths over time are necessary, both to establish the generalizability of our results to other populations (e.g., other majors, non-college graduates) and to provide more nuanced analyses (e.g., effects of demographic variables and other individual differences).